1 Impact of Graupel Class in Microphysics Parameterizations on Bow
Transcript of 1 Impact of Graupel Class in Microphysics Parameterizations on Bow
Impact of Graupel Class in Microphysics Parameterizations on Bow Echo1
Simulations2
Rebecca D. Adams-Selin∗+1, Susan C. van den Heever+, and Richard H. Johnson+3
∗Atmospheric and Environmental Research, Inc.,4
Lexington, Massachusetts5
+Department of Atmospheric Science6
Colorado State University7
Fort Collins, Colorado8
August 20119
(submitted to Monthly Weather Review)10
1Corresponding author address: Rebecca D. Adams-Selin, HQ Air Force Weather Agency 16th WeatherSquadron, 101 Nelson Dr., Offutt AFB, NE, 68113. Email: [email protected]
1
Abstract
In this study, the effect of changes in microphysical cooling rates on bow echo development11
and longevity are examined, through changes to the graupel parameter in a microphysics12
parameterization in the Advanced Research Weather Research and Forecasting (WRF-ARW)13
model. Multiple simulations are performed, testing the sensitivity to numerous graupel size14
distributions, as well as the complete removal of graupel. It is found that size distributions15
with larger and heavier, but fewer, graupel particles result in weaker cold pools due to16
reduced microphysical cooling rates, and therefore weaker mid-level buoyancy and pressure17
perturbations, weaker rear inflow, and a less intense convective updraft that did not tilt18
as required for bowing to occur. Graupel size distributions with more numerous, smaller,19
and lighter particles result in larger microphysical cooling rates, stronger cold pools, and20
stronger mid-level buoyancy and pressure gradients as well as rear inflow; these systems21
develop bowing segments earlier and more often. Simulations without graupel initially are22
weaker, due to limited contributions from cooling by melting from the slowly falling snow.23
However, the system is able to stay more organized with the slightly weaker cold pool, and24
over time the increase in melting snow resulted in a strongly developed system with multiple25
instances of new bowing. Changes in the timing and development of the stratiform rain26
region and convective line, as well as system propagation speed, are also examined.27
2
1. Introduction28
It is well known that a key component to the strength, structure, and longevity of squall29
lines and bow echoes is the cold pool (Rotunno et al. 1988; Weisman et al. 1992, 1993).30
Changes in the rates of cooling by microphysical processes naturally have a large effect31
on the shape and strength of the cold pool, and hence the eventual storm structure and32
development. While the impact of microphysics scheme variations on cold pool and storm33
structure have been noted for supercells (Gilmore et al. 2004; van den Heever and Cotton34
2004, 2007; Dawson et al. 2007, 2008; Snook and Xue 2008; Lerach et al. 2008), few studies35
have looked at the effects on mesoscale convective systems, specifically squall lines and bow36
echoes.37
The cold pool is important not only to the longevity of a squall line system, but also to the38
initiation of any new bowing along the line. A balance among the vorticity generated at the39
front edge of the cold pool, by the environmental wind, and above and below the rear inflow40
into the system can act to ensure the longevity of the system (Weisman 1992). However,41
a temporary, local intensification of the cold pool and its associated vorticity can result in42
a portion of the storm temporarily tilting upshear. The decrease in cold pool temperature43
would also increase the mid-level buoyancy gradient, thereby intensifying the mid-level low44
pressure perturbation; this would also act to tilt the system upshear (Weisman 1993). This45
tilt allows horizontal momentum to be more easily transported to the surface, resulting46
in damaging downburst winds. The combination of additional horizontal momentum and47
a colder cold pool acts to locally increase the speed of the convective line, resulting in48
development of a new bowing segment.49
The microphysical processes in the trailing stratiform region of mesoscale convective50
system were diagnosed by Gallus and Johnson (1995) using a two-dimensional model, and51
by Zhang and Gao (1989), Yang and Houze (1995), and Braun and Houze (1997) with52
three-dimensional models. At initiation, cooling by melting and evaporation takes place just53
behind the convective line, resulting in a cold pool with a distinct head. As the system54
3
matures enhanced melting and evaporation rates spread farther to the rear of the storm,55
forming a large “mound-shaped” cold anomaly. Cooling by sublimation is found throughout56
the base of the stratiform cloud. The microphysics particle distribution observed in a typical57
trailing stratiform region is noted by Smull and Houze (1985) and Biggerstaff and Houze58
(1991). Immediately rearward of the convective line is a region of minimal reflectivity, or the59
“transition zone”. This reflectivity minimum is due to natural sorting by particle fall-speed:60
the hail and larger graupel particles fall out almost immediately close to the convective61
line, while smaller snow particles are advected rearward some distance by front-to-rear flow.62
These smaller particles begin descending in the center of the stratiform region, creating a63
secondary reflectivity maximum.64
Ideally, a simulation of a linear convective system should reproduce these features, as65
otherwise many other features of the system are not correctly simulated. The importance66
of the inclusion of ice in convective squall line simulations has been noted multiple times in67
the literature. The additional intensification of the updraft due to latent heat from freezing68
is required to best simulate the updraft strength (Nicholls 1987). Cooling by melting also69
contributes significantly to the mid-level thermal gradient in the stratiform region (Chen70
and Cotton 1988; Szeto and Cho 1994), and therefore also the mid-level pressure perturba-71
tion and magnitude of the rear inflow (Yang and Houze 1995). These same microphysical72
processes could therefore also be expected to have an impact on the initiation on new bow-73
ing development, through changes in the mid-level thermal and pressure gradients in the74
stratiform region. However, there have been no studies linking the two.75
Thus, a cloud-resolving model is used to examine the impacts of microphysics specifically76
on bow echoes. As ice processes have been shown to have such a large effect on these77
mid-level thermal and pressure gradients, several characteristics of frozen hydrometeors are78
varied within this model. The sensitivity of new bowing frequency and intensity to changes79
in microphysical heating and cooling rates will be investigated, particularly through their80
relation to cold pool strength and depth, the mid-level thermal and pressure gradients within81
4
the system,and the stratiform precipitation microphysical structure.82
Section 2 describes the experiment and model design, as well as the specific microphysics83
variations. A review of the case study being simulated is presented in Section 3. Sections 484
and 5 contain results from the two different sensitivity studies performed, and Section 6 the85
conclusions.86
2. Model description and setup87
a. Model and experiment design88
The Advanced Research Weather Research and Forecasting (WRF-ARW) model is widely89
recognized as a reliable tool for real-data simulations of mesoscale phenomena. Here version90
3.2 was used to simulate an isolated bow echo case over Oklahoma on 13 March 2003. For91
this case study, the initial conditions were provided by the National Center for Environmen-92
tal Prediction (NCEP) Final Operational Global Analysis data, at 1o by 1o horizontal and93
six-hour temporal resolution. The model was initialized at 1200 UTC on 12 March 2003, and94
run for 36 hours to 0000 UTC 14 March 2003. A horizontal resolution of 3 km was used with95
35 vertical levels; the domain is shown in Fig. 1. The vertical levels were stretched, with96
increasing resolution in the lower levels. Parameterization schemes other than microphysics97
include the Mellor-Yamada-Janjic boundary layer scheme (Janjic 1994), the Noah land sur-98
face model, Rapid Radiative Transfer Model longwave radiation scheme, and the Goddard99
shortwave radiation scheme. No convective parameterization was used. This combination, in100
addition to the set of microphysics schemes, was chosen after considerable experimentation101
as they produced the most realistic convective systems as compared to observations.102
A set of idealized simulations was run to isolate microphysical effects from synoptic103
influences, as well as to allow these results to be generalized to more than one specific104
case study. Higher vertical and horizontal resolution was also used to resolve finer features105
within the system. The 0000 UTC 13 March 2003 KOUN sounding (Fig. 2) provided the106
homogenous initial conditions. Other than extrapolation of moisture data at upper levels107
and minimal smoothing to remove instabilities, the sounding was unmodified. The model108
5
horizontal resolution was 1 km, with 72 stretched vertical levels. In total, the domain109
extended 600 km in the x direction, 400 km in the y direction, and 18 km vertically. No110
parameterizations, other than microphysics, were used. The simulations were run for six111
hours, as this was the approximate lifetime of the observed bow echo. The “cold pool-dam112
break” initialization scheme, originally created for the CM1 model (Bryan and Fritsch 2002),113
was modified for these simulations. A “cold dam” of air was created by decreasing the initial114
potential temperature in the domain, from 0 to 200 km in the x-direction, and 50 to 350115
km in the y-direction. The magnitude of the perturbation was 6 K at the surface, and116
linearly decreased until reaching 0 K 2.5 km aloft. Upon simulation start, the “dam” of117
cold air would break, surging forward as a gravity current. Air in advance of the gravity118
current would be forced upward, initiating convective activity. This initialization method119
was specifically chosen due to the low-level stable layer in the initial sounding (Fig. 2). This120
stable layer acted as a cap and required a larger amount of forcing to be overcome than121
would be provided by a warm bubble initialization.122
b. Model microphysics123
The WRF Single-Moment and Double-Moment 5- and 6-class microphysics schemes124
(Hong et al. 2004; Hong and Lim 2006; Lim and Hong 2010) were originally based upon the125
techniques used in Lin et al. (1983) and Rutledge and Hobbs (1983). The 5-class schemes126
contain explicit classes for water vapor, cloud water, raindrops, cloud ice, and snow; the127
6-class schemes add graupel. All of these schemes utilize an inverse exponential Marshall-128
Palmer size distribution for snow and, for the 6-class, graupel:129
nx(Dx)dDx = n0xexp(−λxDx)dDx (1)
where x is the microphysics class, nx(Dx)dDx is the number of particles per cubic meter130
with diameters between Dx and Dx + dDx, n0x is the distribution intercept, and λx is the131
slope. The distribution intercept is set to a constant value prior to processing. This is true132
even for WDM5 and 6, as they only explicitly calculate the second moment of concentration133
6
for the raindrop and cloud water distributions. The slope is a diagnosed value, defined as:134
λx =
(πρxn0x
ρqx
)0.25
(2)
where ρx is the pre-assigned particle density, ρ is the local air density, and qx is the prognostic135
particle mixing ratio.136
From (1) and (2), it can be seen that the size distribution of a particle is a function of137
the intercept, n0x, and the slope, λx, which is a function of the particle density. Due to the138
inverse nature of the distribution, a large intercept will result in a smaller mean particle139
size, and vice versa. Within this experiment, graupel particles have a smaller mean particle140
size (larger intercept) and are less dense, while hail particles have a larger mean particle size141
(smaller intercept) and are more dense, but in observed data there is quite a bit of overlap142
between the two. From Gilmore et al. (2004), previous observations of the intercept value143
for hail and graupel ranged from 102 m−4 for large hail, to 1010 m−4 for extremely small144
graupel (Cheng et al. 1985, Dennis et al. 1971; Federer and Waldvogel 1975; Spahn 1976;145
Knight et al. 1982). Observations of graupel density ranged from 50 to 890 kg m−3; hail146
density varied from 700 to 900 kg m−3 (Pruppacher and Klett 1978). In the first portion of147
this experiment, model runs are performed using the WSM6 scheme, but covering this range148
of variables. The most “hail-like” particle has an intercept equal to 4x102 m−4 and a density149
of 900 kg m−3. The most “graupel-like” particle has an intercept equal to 4x106 m−4 and150
a density of 300 kg m−3. A plot of the graupel particle size distributions given the chosen151
intercepts and densities is shown in Fig. 3.152
As discussed, it is evident that a decrease in the particle intercept yields an overall153
increase in the number of large particles while decreasing the number of small particles.154
This change in mean particle size will increase the mean terminal velocity. This will result in155
less dwell time in the downdraft, and therefore less time for melting and evaporation. Larger156
particles also have less surface area-to-volume ratio, which further reduces the melting and157
evaporation rates (van den Heever and Cotton 2004). A change in density alone has a much158
smaller effect on these rates, but it will act to slightly decrease the number of very large159
7
particles. This would thereby also decrease the mean terminal velocity, allowing slightly160
more time for melting and evaporation.161
In the second half of the experiment, a comparison of the 5- and 6-class schemes is162
used to examine the importance of graupel as a class in simulating this type of convective163
system. These will be termed the “graupel” (6-class) and “no-graupel” (5-class) simulations.164
A pair of “no-graupel” and “graupel” simulations were run using the WRF Double-Moment165
5- and 6-class schemes, and another pair using the WRF Single-Moment 5- and 6-class166
schemes. Similar results are obtained between these pairs, with one notable exception. The167
simulated reflectivity in the single-moment simulations is less intense than was observed in168
the convective line. This deficiency is noted throughout each single-moment simulation, while169
the WRF double-moment schemes do not have such a problem. However the double-moment170
schemes instead produce too intense stratiform precipitation. Examination of the differences171
between the single- and double-moment schemes will be reserved for a later study. For this172
article, the case-study model runs using the “graupel” and “no-graupel” WDM schemes are173
compared as these runs better simulated what was observed. However, for the idealized174
“graupel” and “no-graupel” simulations, the WSM schemes are chosen for the discussion175
due to their reduced complexity, thus allowing for a more straightforward comparison of176
microphysics composition and cooling rates.177
A full list of the microphysics schemes and associated intercept and density parameters178
used in each simulation is provided in Table 1.179
3. Case review180
Convection first initiated at 0215 UTC in north-central Oklahoma, and a convective181
line aligned southwest to northeast approximately 250 km in length formed by 0315 UTC182
(Fig. 4a). The system moved to the southeast at a speed of approximately 11 m s−1. A trail-183
ing stratiform precipitation region appeared at 0500 UTC, and began to develop rearward.184
At 0600 UTC (Fig. 4b) the center of the convective line bowed out; stratiform precipitation185
filled in behind the line as it bowed. A secondary reflectivity maximum formed in the strati-186
8
form region at 0700 UTC, separated from the convective line by a transition zone. The bow187
echo continued to increase in size until 0800 UTC (Fig. 4c). At this point the convective188
line started to dissipate. The gust front associated with the decaying convective line, and189
the stratiform precipitation, continued to propagate southeastward over the next four hours.190
This eventually reformed into another mesoscale convective system in Louisiana.191
4. Results from hail-graupel comparison192
a. Case study simulations193
The spectrum of results between the “hail-like” simulation and the “graupel-like” sim-194
ulation can be summarized by comparing the hail-like and graupel-like runs (Fig. 5). Both195
the hail-like and graupel-like simulations initiated convection somewhat differently than ob-196
served: an isolated convective cell developed in southwestern Oklahoma at 0300 UTC and197
moved southeast. By 0600 UTC, the convection in the hail-like run was cellular and appeared198
disorganized (Fig. 5a); a bow-shaped convective line formed in south-central Oklahoma in199
the graupel-like run (Fig. 5b). The linear convection to the northeast of the initial system200
will not be considered here as it has no correspondence to observations. As the hail-like201
simulation progressed, it was simply unable to reproduce any of the typical features of a202
bow echo, particularly any sort of convective line (0800 UTC, Fig. 5c). The hail-like sys-203
tem has less overall precipitation, both in intensity and area, than the graupel-like system.204
The system speed was also slower. At 0800 UTC the graupel-like simulation has further205
developed its bow shape as well as a stratiform precipitation region (Fig. 5d). Both systems206
began to dissipate at 0900 UTC (not shown), somewhat later than observed. However, the207
graupel-like system still appeared more organized, retaining its bow shape and some con-208
vective intensity while dissipating; this was also true of the observed system (Fig. 4c). The209
reason for this will become clear after further examination.210
Figures 6, 7, and 8 show cross-sections of each storm system taken across the convective211
line at 0600 and 0800 UTC (see thick black lines in Fig. 5 for exact placement). In the hail-212
like simulation at 0600 UTC, cloud ice and water, typically found in the updraft, and hail,213
9
typically found in the downdraft, were located very close together (Fig. 6a). From this one214
can surmise the hail particles were too large to be sustained aloft in the convective updraft215
and fell out of the hail-like system almost immediately upon formation. Cloud ice crystal216
concentrations were also much higher in the upper levels of the storm; these ice crystals217
were not being scavenged by numerous semi-melted graupel particles as in the graupel-like218
simulation. The convective updraft in the hail-like simulation, which was almost entirely219
above 500 hPa (Fig. 8a), was nevertheless very upright and strong. This was due to the220
large concentration of condensation, freezing, and deposition processes immediately within221
and next to the updraft. From Fig. 7a, it is evident that the hail particles were so large, and222
fell so quickly, that significant cooling by melting did not occur until almost reaching the223
surface. Hence, there was also limited cooling by evaporation of raindrops. As a result, the224
cold pool was very shallow, and of weaker intensity (Fig. 6a). All cooling at this time was225
concentrated almost immediately behind the convective line; the cold pool was very narrow226
as well. Due to the weaker intensity, there was a gentle upward slope of potential temperature227
contours at the front of the hail-like system (Fig. 6a). Because of this, the forcing of the air228
upward by the leading edge of the gently-sloped cold pool was not as abrupt, reducing the229
intensity of the low-level updraft (Fig. 8a).230
Additionally, it is known the buoyancy gradient in a system with respect to height can231
be related to the associated pressure perturbation through232
1
ρ0
∇2p′ = −∇ · (~v · ∇~v) +∂B
∂z(3)
where p′ is the perturbation pressure, ~v is the horizontal wind vector, and B is buoyancy,233
defined as gθ′v/θv. The mid-level buoyancy gradient is typically positive, due to the warmer234
updraft overlying the cold pool. In the hail-like case, this gradient was still positive, but235
smaller, due to the warmer cold pool. Thus, (3) yielded a positive but small ∇2p′, and236
therefore a weak mid-level low pressure perturbation (Fig. 8a). Because this perturbation237
was so weak, no extra source of vorticity from the rear inflow within the system was required238
for the updraft to remain upright instead of tilted. Minimal rear-to-front flow was present239
10
(not shown).240
In the graupel-like system, precipitation was advected rearward farther from the convec-241
tive line, as evidenced by the larger, 70-km wide stratiform region (Fig. 6b). There were large242
concentrations of graupel and snow both closer to the convective line and farther behind the243
system. Because of the smaller size of the graupel particles, they were more easily held aloft244
and advected rearward. From Fig. 7b, high rates of deposition, melting, and evaporation of245
the falling graupel and snow out of the stratiform precipitation region were evident. Due to246
the smaller size of these particles, and hence the larger surface-area-to-volume ratios, they247
underwent sublimation, melting, and evaporation quickly; the small particles melted almost248
immediately after falling below the melting level. These processes were also enhanced by249
the longer dwell time aloft due to their slower fall speeds. Therefore the cooling rates in250
graupel-like were much larger, and also were spread over a much larger area than in the hail-251
like run. The heightened cooling rates resulted in a more intense, deeper, and larger cold252
pool (Fig. 6b), as discussed earlier. Specifically, the cold pool in the graupel-like simulation253
was 4 K colder than the hail-like simulation cold pool. Because hydrometeor particles were254
being advected farther rearward, freezing and deposition rates were higher across the upper255
levels of the stratiform region. The latent heat released by these processes helped create a256
stronger and larger stratiform updraft. Additionally, the wider, colder cold pool allowed for257
a stronger buoyancy gradient with respect to height, increasing the mid-level low pressure258
gradient, and therefore also increasing the convective updraft intensity and its rearward tilt259
(Fig. 8b). Rear-to-front flow in the low- and mid-levels of the system to the rear of the260
convective line also strengthened (not shown). Finally, gravity current theory suggests a261
more intense cold pool would result in a faster-moving convective system; this was indeed262
the case.263
At 0800 UTC in the hail-like simulation, the hail and rain had begun falling into the264
updraft (Figs. 6c, 8c). The system appeared similar to a dying pulse-type thunderstorm.265
The cold pool and mid-level pressure perturbation were both still minimal, almost no rear266
11
inflow was present, and the low-level convective updraft was weak and still upright; no267
bowing segments were ever produced. There was almost no rear-to-front flow within the268
system. The deepest part of the cold pool had shifted slightly rearward of the convective269
line, but peak cooling rates (not shown) still remained much closer to it than in the graupel-270
like run. In that simulation, the stratiform region was still significantly larger than in the271
hail-like simulation; the cold pool was as well (Fig. 6d). However, peak cooling rates had272
shifted farther rearward into the area below the middle of the stratiform region. The deepest273
part of the cold pool was approximately 50 km behind the convective line. Meanwhile the274
mid-level low pressure perturbation associated with this system had grown stronger and275
shifted rearward (Fig. 8d). The rear inflow immediately behind this pressure perturbation,276
at the very rear of the system, was still strong. However, the rear-to-front flow forward of277
the pressure perturbation had weakened significantly, resulting in no additional sources of278
vorticity to help the convective updraft, at the front of the system, remain upright. As a279
result the low-level convective updraft was tilted even more rearward over the cold pool.280
This proved to be too extreme of a tilt, as the graupel-like system began dissipating shortly281
after this by 0900 UTC.282
b. Idealized cases283
A set of idealized simulations, as described above, were run to determine if the results seen284
in the case study were an artifact of that specific case study, or if they could be generalized285
to multiple situations. They were also run at a higher horizontal (1 km versus 3 km) and286
vertical (72 levels instead of 35) grid spacing to better resolve the storm structure. The287
major difference between the idealized and case study simulations was the organization and288
intensity of the hail-like simulation. Two hours into the idealized simulations (Fig. 9a,b),289
after convection was able to initiate and organize, the hail-like system had an equally intense290
but much narrower leading convective line, with a trailing stratiform region similar in size to291
the graupel-like system, but also less intense. Both systems began to bow in a similar manner292
about three hours into the simulation. At 4.5 hours, the systems were fully mature, and the293
12
hail-like system had actually developed another bowing segment (Figs. 9c,d). Both systems294
had about equally sized stratiform precipitation regions. The reason for the idealized hail-like295
system’s increased organization and intensity, as compared to the case study simulations,296
is likely due to the lack of radiation cooling. As the case study simulation progressed in297
time, the sun set and the boundary layer naturally cooled (bowing developed in the graupel-298
like case study simulation at 0000 LST). This did not occur in the idealized simulation,299
so its system was able to continually ingest warm, unstable air in the lower levels. This300
helped intensify the updraft and the front-to-rear motion of all precipitation particles, not301
just the hailstones, creating a somewhat larger stratiform precipitation region. This allowed302
for additional cooling by melting and evaporation and therefore a slightly stronger cold303
pool. The associated buoyancy gradient, and stronger, more elevated rear inflow was intense304
enough to eventually create a bowing convective line. Additionally, the very strong forcing305
generated by the initial 6 K perturbation density current designed to initialize convection306
also acted to organize that initial convection. The convection in the case study had little307
synoptic forcing, and thus had to rely on internal dynamics to organize into a bowing system;308
we have already noted the internal dynamics in the hail-like case study simulation were weak.309
Despite these differences, however, the most important results seen in the case study310
simulations were also evident in these idealized simulations. In the hail-like system, the311
hail still fell out of the system very close to the updraft. Its very fast fall speed again312
resulted in minimal cooling by melting and evaporation, a minimal cold pool, and thus a313
slower system speed. The vertical buoyancy gradient and related mid-level low pressure314
perturbation were also less intense, leading initially to less rear inflow and a correspondingly315
weaker but upright lower-level convective updraft that did not tilt. While bowing did occur,316
it was over a smaller area than the graupel-like simulation, and occurred approximately half317
an hour later. Meanwhile, in the graupel-like simulation the system was more organized.318
Due to the slower fallspeeds of the graupel particles, they not only were advected farther319
away creating a larger stratiform region, but also melted quickly thereby generating strong320
13
cooling rates. Hence, the cold pool in the graupel-like simulation was intense and deep, and321
the speed of the system, as well as the rear-to-front flow within it, was correspondingly fast.322
The sharp leading edge of the cold pool helped to intensify the low-level convective updraft,323
as did the more intense mid-level pressure perturbation. This led to a temporary tilting of324
the low-level convective updraft and a bowing segment forming earlier, and covering a wider325
area.326
Thus, in both the idealized and the case study simulations, changing the graupel pa-327
rameter to take on more hail-like characteristics acted to diminish the microphysical cooling328
within the system. This in turn decreased the strength of the cold pool, and weakened329
the low-level convective updraft both by decreasing the mid-level pressure perturbation and330
flattening the slope of the potential temperature contours at the front edge of the cold pool.331
A weaker mid-level pressure perturbation also resulted in the low-level convective updraft332
never tilting upshear. Hence bowing segments did not develop as quickly or over as large of333
an area - or develop at all without additional forcing, as in the case study.334
5. Results from removal of graupel class335
The WRF Double-Moment 5-class and 6-class microphysics schemes, and the WRF336
Single-Moment 5-class and 6-class microphysics schemes, are identical except for the in-337
clusion of graupel as a class in the 6-class schemes (Lim and Hong 2010). Simulations using338
both the 5- and 6-class schemes (“no-graupel” and “graupel” simulations) were run to exam-339
ine the importance of graupel as a class, as hinted at in the above hail-graupel comparisons.340
These comparisons will continue to explore the effect of changes in microphysical cooling341
rates on bowing initiation, as naturally removing graupel as a class will affect these rates.342
Additionally, many operational models that are intended to forecast convection, such as the343
Weather Research and Forecasting Nonhydrostatic Mesoscale Model (WRF-NMM) and the344
Regional Global Environmental Multiscale (Regional GEM), all use no more than five classes345
within their microphysics schemes due to computing constraints. Thus an exploration of the346
differences that can be expected when simulating this type of convection without graupel347
14
would be beneficial for operational purposes.348
a. Case study349
Figure 10 shows the simulated model reflectivity for the case study 5-class (“no-gruapel”)350
and 6-class (“graupel”) runs. During initiation (0445 UTC), the no-graupel simulation gener-351
ated unrealistically large stratiform regions both in advance of and behind the still developing352
convective line (Fig. 10a). This was due to the widespread concentrations of slowly falling353
snow crystals in the upper levels (Fig. 11a). At this time, the no-graupel system’s stratiform354
precipitation and minimal melting and evaporative cooling remained entirely aloft, resulting355
in an almost non-existent cold pool (Fig. 11a). At 0530 UTC, precipitation reached the356
ground, and the convective line continued to develop and was fully in place by 0600 UTC357
(Figs. 10b, 11c). The atmospheric layers beneath the stratiform region were saturated as358
the system reached maturity, and the stratiform precipitation extended completely to the359
surface. While there were more, smaller snow crystals to melt and evaporate theoretically360
leading to a stronger cold pool, these saturated layers did not allow as much evaporation to361
take place. Thus, the cold pool in the no-graupel simulation was significantly shallower and362
warmer than in the graupel simulation, as will be seen later. By 0730 UTC, however, the no-363
graupel simulation gained convective intensity and began bowing (Fig. 10c). The system was364
well organized, with a tightly defined convective line and less intense but extensive stratiform365
region that by this point was entirely rearward of the convective line. There was additionally366
a secondary maximum of reflectivity in the stratiform region, separated from the convective367
line by a transition zone, which matched observations well. The area beneath the stratiform368
precipitation region was no longer saturated by this time (Fig. 11e) thus allowing for a more369
developed cold pool. Yet, because the cold pool was initially warmer, the no-graupel system370
moved relatively slowly. Possibly this speed, and the resulting storm-relative environmental371
wind shear, was optimal as it allowed the vorticity generated by the cold pool and that by372
the environmental wind shear to balance each other well. The initially warmer cold pool373
also meant a less intense mid-level buoyancy gradient and low pressure perturbation, further374
15
helping the convective updraft to remain upright initially, but by 0730 UTC the mid-level375
pressure perturbation had intensified (Fig. 12a). The low-level convective updraft was able376
to tilt rearward over the cold pool temporarily, allowing a bowing segment to form; the377
increase in system rear inflow strengthened the resulting surface winds. The overall system378
became more intense as time progressed, and a more organized system was created that379
better matched observations. At 0900 UTC, the system began to dissipate, although the380
convective line was still fairly intense (greater than 45 dBZ).381
The system in the graupel simulation quickly developed intense convection by 0445 UTC382
(Fig. 10d). At this point the convection was cellular, with minimal stratiform precipitation.383
The larger graupel hydrometeors were not as mobile, and stayed closer to the convective384
updraft. (Fig. 11b). The cold pool was still very weak at this point, but was larger than the385
corresponding time in the no-graupel simulation. Cooling rates due to melting graupel and386
evaporating rain were approximately equal at this point (not shown); the sub-cloud layer387
was sub-saturated. By 0600 UTC, the system developed mature mesoscale convective system388
features, such as a convective line and intense stratiform precipitation region (Fig. 10e). The389
stratiform region was composed of large concentrations of graupel and snow aloft, but these390
concentrations were not as large, both in quantity and area, as the snow concentrations in the391
no-graupel run at the same time (Fig. 11d). The cold pool, created by both melting graupel392
and evaporating rain falling through a still sub-saturated sub-cloud layer, was much larger,393
deeper, and cooler than the no-graupel system (6 K cooler at its most intense point). The394
cold pool at this point also had a distinct head, as the cooling maximum in the graupel system395
was closer to the convective line due to the graupel not spreading as far from the updraft.396
At 0730 UTC, the stratiform precipitation area was much less organized (Fig. 10f), with397
diminished amounts of snow and graupel particles aloft (Fig. 11f). Examination of vertical398
wind speeds revealed that the low-level convective updraft at the front edge of the cold pool399
had moved more quickly than the upper-level updraft associated with the stratiform region400
(Fig. 12b). By this point, the disparate speeds forced the disorganization of the system.401
16
The cold pool developed a mound shape with the deepest point beneath the stratiform402
precipitation region, as the main source of cooling shifted to beneath the stratiform region403
(Fig. 11f). The mid-level low pressure perturbation had also remained strong and shifted404
farther rearward, further helping to tilt the updraft (Fig. 12b), as did the spreading of the405
rear inflow along the surface up to 50 km behind the convective line (not shown). At this406
time, the graupel system remained intense, but was disorganized; after 0800 UTC the system407
lost its convective line and began to slowly dissipate.408
b. Idealized cases409
Two idealized simulations were also run using the WSM5 (no-graupel) and WSM6 (grau-410
pel) microphysics schemes, again to determine if the differences between the 5- and 6-class411
schemes seen in the case study simulations discussed above were partially due to synoptic412
heterogeneities, or if they were entirely microphysics-driven. Figure 13 displays the sim-413
ulated composite reflectivity calculated for the two simulations. The similarities between414
these idealized systems and those produced in the case study simulations are striking. The415
large precipitation region ahead of the convective line during the initial hours of its forma-416
tion in the no-graupel system was still due to the large concentration of slowly falling, easily417
advected snow produced (Fig. 13a, b). Because the melting and evaporation of the snow418
initially occurred aloft, the cold pool was minimal, and the system correspondingly slow,419
until the snow fell below the melting level and began to fall as rain. As in the case-study420
no-graupel simulation, the cooling by melting and evaporation eventually intensified the cold421
pool enough to allow the convective line to bow (Fig. 13c), through a local intensification of422
the buoyancy gradient, mid-level low pressure perturbation, and system rear inflow behind423
the convective line.424
As expected, the stratiform precipitation region was much smaller in the graupel case,425
due to the relatively larger, faster-falling graupel particles (Fig. 13d). Cooling rates were426
stronger in the graupel case, due to the melting and subsequent evaporation of the faster-427
falling graupel spread throughout the lower levels. However, the graupel fallspeeds were not428
17
too fast to preclude large melting and evaporation rates before reaching the surface. The cold429
pool in the graupel simulation was therefore much stronger. Again, the graupel fell closer to430
the convective line, making it more intense (Fig. 13e) and thus the peak cooling also remained431
closer to the convective line. This resulted in the cold pool having a shaper, “head”-shaped432
leading edge at this time. The steep leading edge of the cold pool helped to intensify the433
leading updraft, while the increased intensity of the mid-level pressure perturbation due to434
the stronger buoyancy gradient resulted in a more favorably tilted updraft. Thus, the system435
rear-to-front flow was stronger, and the system began to bow half an hour earlier. However,436
the peak cooling rates eventually shifted rearward as in the case study, resulting in a more437
gentle slope of the potential temperature contours at the front edge of the cold pool, and438
a very strong mid-level low pressure perturbation that had also shifted rearward. The rear439
inflow began to spread along the surface far behind the convective line. All of these things440
acted to tilt the low-level updraft too far rearward over the cold pool, weakening the system.441
Thus, the main differences between the no-graupel and graupel systems observed in442
the case study were also observed in the idealized simulation. Specifically, the no-graupel443
simulation cold pool transitioned from relatively warmer to cooler as the slower-falling snow444
finally began to melt; the system itself became more organized as the simulation progressed.445
Meanwhile the graupel simulation cold pool was very strong from shortly after initialization,446
due to the large cooling rates produced by melting graupel and evaporating rain spread447
throughout the lower levels. However, the cooling rates shifted rearward as the simulation448
progressed; the deepest part of the cold pool shifted rearward as well, as did the mid-level449
low pressure perturbation. The low-level convective updraft was tilted too far rearward over450
this intense cold pool, and the system therefore began to weaken earlier than the 5-class451
simulation.452
6. Conclusions453
Multiple WRF convective simulations were conducted to examine the effect of changes in454
microphysical cooling rates on bow echo generation and longevity, specifically through the455
18
connection between the mid-level buoyancy gradient and low-level convective updraft tilt.456
Simulations were performed over a spectrum of graupel class parameters in a microphysics457
parameterization scheme, rendering the class larger, more dense and “hail-like”, or smaller,458
lighter and “graupel-like”. Additional simulations looked at the effects of removing graupel459
completely. Both case study and idealized simulations were performed to generalize these460
results beyond one specific case.461
The simulations with a larger, more dense “hail-like” graupel class had these particles fall462
out of the updraft almost immediately, close to the convective line, allowing little melting or463
evaporation. This resulted in a minimal stratiform precipitation region, little organization,464
and reduced convective intensity. Because of this, there were fewer feedbacks to updraft465
intensity, such as a weaker and more shallow cold pool, very reduced rear inflow, and a466
weaker mid-level low pressure perturbation. In the case study, there was little convection467
and no bowing. The idealized case did exhibit bowing behavior, although later in time and468
over a smaller area than the idealized graupel-like case. It is likely the lack of radiation469
cooling sustained the system by providing it with a continuous source of warm, moist air at470
lower levels. This resulted in a slightly more intense updraft, providing the hailstones with471
enough rearward momentum to avoid falling directly through the updraft. The initiation472
method also provided additional organization and intensity at the start of convection.473
Meanwhile, the idealized and case study simulations with a smaller, less dense “graupel-474
like” graupel class had a mixture of small snow crystals and larger graupel particles that475
were slower to fall, creating a wide stratiform precipitation region. This allowed for more476
melting and evaporation, and yielded a wider, deeper, and stronger cold pool. The mid-level477
low pressure perturbation associated with the system was stronger, strengthening the low-478
level convective updraft and aiding it in temporarily tilting rearward. This helped generate479
bowing segments earlier and over a larger area than the hail-like simulations. The cold pool480
retained its “head” shape longer. However, the rearward advection of graupel produced481
an unrealistically intense stratiform precipitation region, with no transition zone reflectivity482
19
minimum.483
The various cloud and ice particles generated by the no-graupel (5-class) simulations re-484
mained entirely aloft for some time after initiation as the large amounts of snow generated485
fell very slowly. Each stratiform precipitation region only reached the surface as the system486
became vertically saturated. This saturation allowed for little evaporation; hence the cold487
pool was less intense and the system was slower. Due to the weaker buoyancy gradient and488
mid-level low pressure perturbation, the low-level convective updraft was not as strong, the489
rear inflow weaker, and the convective line initially not as intense as the graupel simula-490
tions (6-class). However, this slower speed allowed the no-graupel systems to remain more491
organized. Over time the cooling by evaporation rates increased, intensifying the mid-level492
pressure perturbation and temporarily tilting the low-level updraft, developing a bow echo.493
Due to the increased horizontal resolution, the idealized no-graupel simulation was also able494
to better segregate its particle sizes in the stratiform precipitation region, creating a transi-495
tion zone and secondary reflectivity maximum. The graupel simulations initially developed a496
stronger cold pool, as the sub-cloud layer was not saturated. The stronger cold pool resulted497
in bowing developing faster. However, as each simulation progressed the cold pool became498
too strong. The fast speed of the system, coupled with the stronger mid-level lower pressure499
gradient and the rear inflow spreading to the surface far behind the convective line, helped500
tilt the updraft too far over the cold pool. The system became disorganized more quickly,501
approximately an hour earlier than the no graupel simulations.502
Thus, it was found that internal variations in the WRF Single- and Double-Moment503
microphysics schemes, specifically regarding the graupel class, had a large effect on the504
structure, strength, and longevity of the simulated convective system. Changes in the mi-505
crophysical cooling rates affected the tilt of the low-level convective updraft through the506
mid-level low pressure perturbation, affecting the timing, size, and existence of bowing.507
These results were noted in both case study and idealized simulations, ensuring they were508
not due to synoptic heterogeneities.509
20
The similarity of these results to those obtained by Weisman (1993), which used simu-510
lations with no ice, is noted. Obviously frozen particles and their associated microphysical511
processes are not required for bowing development. However, in these simulations it was512
found that the cooling produced by melting can play a significant role in creating a cold513
pool strong enough to initiate bowing. A delay in the onset of cooling by melting, as in the514
no-graupel simulations, helped the system retain its organization for a longer period of time.515
Extremely low melting rates, such as in the hail case, resulted in much weaker systems that516
in some cases did not bow. Thus the impacts of melting rates should not be discounted in517
future bowing simulations.518
While this study has examined sensitivities to model microphysical schemes, such vari-519
ations to the microphysics can occur in reality. For example, introduction of dust, lofted520
by storm outflow, into a system can drastically modify the concentration of both cloud con-521
densation and ice-forming nuclei, as was seen in Twohy et al. (2010). This naturally results522
in significantly different size distributions of all microphysical particles, noted by van den523
Heever et al. (2006) and Storer et al. (2010) for deep convection; this particularly affects the524
heating and cooling rates as discussed here. There are many further environmental factors525
that could result in similar microphysical changes, such as variations in nearby temperature,526
moisture, or shear profiles, so these results are of note. Future work includes using WSR-88D527
radar data to more fully compare the model simulations to observations. The differences be-528
tween multi-moment schemes, particularly the single- and double-moment schemes noted at529
the start of this work, also need to be more fully explored.530
7. Acknowledgments531
This research was supported by National Science Foundation Grant ATM-0500061, and532
conducted under the Cooperative Research Data Agreement between the Air Force Weather533
Agency and Atmospheric Environmental Research, Inc. Susan van den Heever is supported534
by the NSF under Grant ATM-0820557. Computing resources were provided by the Navy535
Department of Defense Supercomputing Resource Center (Navy DSRC) and the Army Re-536
21
search Laboratory Department of Defense Supercomputing Resource Center (ARL DSRC),537
which are sponsored by the DoD High Performance Computing Modernization Program.538
The authors would like to thank Morris Weisman, George Bryan, and Kevin Manning of539
NCAR, and Steven Rutledge of CSU, for discussions regarding this work and assistance in540
using the WRF model.541
22
REFERENCES
Biggerstaff, M. I. and R. A. Houze, 1991: Kinematic and precipitation structure of the 10-11542
June 1985 squall line. Mon. Wea. Rev., 119, 3034–3065.543
Braun, S. A. and R. A. Houze Jr. 1997: The evolution of the 10-11 June 1985 PRE-STORM544
squall line: Initiation, development of rear inflow, and dissipation. Mon. Wea. Rev.,545
125, 478–504.546
Bryan, G. H. and J. M. Fritsch, 2002: A benchmark simulation for moist nonhydrostatic547
numerical models. Mon. Wea. Rev., 130, 2917–2928.548
Cheng, L. M. English, and R. Wong, 1985: Hailstone size distributions and their relationship549
to storm thermodynamics. J. Climate App. Meteor., 24, 1059–1067.550
Dennis, A. S. P. L. Smith Jr. G. A. P. Peterson, and R. D. McNeil, 1971: Hailstone size dis-551
tributions and equivalent radar reflectivity factors computed from hailstone momentum552
records. J. App. Meteor., 10, 79–85.553
Dawson, D. T. II, M. Xue, and J. A. Milbrandt, 2008: Improvements in the treatment554
of evaporation and melting in multi-moment versus single-moment bulk microphysics:555
Results from numerical simulations of the 3 May 1999 Oklahoma tornadic storms. 24th556
Conf. on Severe Local Storms, Savannah, Georgia, Amer. Meteor. Soc. 17B.4.557
Dawson, D. T. II, M. Xue, J. A. Milbrandt, M. K. Yau, and G. Zhang, 2007: Impact of558
multi-moment microphysics and model resolution on predicted cold pool and reflectivity559
intensity and structures in the Oklahoma tornadic supercell storms of 3 May 1999. 22nd560
Conf. Wea. Anal. Forecasting/18th Conf. Num. Wea. Pred., Salt Lake City, Utah, Amer.561
Meteor. Soc. CD-ROM 10B.2.562
23
Federer, B. and A. Waldvogel, 1975: Hail and raindrop size distributions from a Swiss563
multicell storm. J. App. Meteor., 14, 91–97.564
Gallus, W. A. Jr. and R. H. Johnson, 1995: The dynamics of circulations within the trailing565
stratiform regions of squall lines: Part I: The 10-11 June PRE-STORM system. J.566
Atmos. Sci., 52, 2161–2187.567
Gilmore, M. S. J. M. Straka, and E. N. Rasmussen, 2004: Precipitation uncertainty due568
to variations in precipitation particle parameters within a simple microphysics scheme.569
Mon. Wea. Rev., 132, 2610–2627.570
Hong, S.Y. J. Dudhia, S.H. Chen, 2004: A revised approach to ice microphysical processes571
for the bulk parameterization of clouds and precipitation. Mon. Wea. Rev., 132, 103–572
120.573
Hong, S.Y. and J.O. J. Lim, 2006: The WRF single-moment 6-class microphysics scheme574
(WSM6). J. Korean. Meteor. Soc., 42, 129–151.575
Janjic, Z. 1994: The step-mountain eta coordinate model: Further developments of the576
convection, viscous sublayer, and turbulence closure schemes. Mon. Wea. Rev., 122,577
927–945.578
Knight, C. A. W. A. Cooper, D. W. Breed, I. R. Paluch, P. L. Smith, and G. Valie, 1982:579
Microphysics. Hailstorms of the Central High Plains, C. Knight and P. Squires, Eds.580
Vol. 1, Colorado Associated University Press, 151–193.581
Lerach, D. G. B. J. Gaudet, and W. R. Cotton, 2008: Idealized simulations of aerosol influ-582
ences on tornadogenesis. Geophys. Res. Lett., 35, L23806, doi:10.1029/2008GL035617.583
24
Lim, K.S. S. and S.Y. Hong, 2010: Development of an effective double-moment cloud mi-584
crophysics scheme with prognostic cloud condensation nuclei (CCN) for weather and585
climate models. Mon. Wea. Rev., 138, 1587–1612.586
Lin, Y.L. R. D. Farley, and H. D. Orville, 1983: Bulk parameterization of the snow field in587
a cloud model. J. Climate App. Meteor., 22, 1065–1092.588
Pruppacher, H. R. and J. D. Klett, 1978: Microphysics of Clouds and Precipitation. D.589
Reidel, 714 pp.590
Rotunno, R. J. B. Klemp, M. L. Weisman, 1988: A theory for strong, long-lived squall lines.591
J. Atmos. Sci., 45, 463–485.592
Rutledge, S. A. and P. V. Hobbs, 1983: The mesoscale and microscale structure and organi-593
zation of clouds and precipitation in mid-latitude cyclones. Part VIII: A model for the594
“seeder feeder” process in warm-frontal rainbands. J. Atmos. Sci., 40, 1185–1206.595
Smull, B. F. and R. A. Houze, Jr. 1985: A midlatitude squall line with a trailing region of596
stratiform rain: Radar and satellite observations. Mon. Wea. Rev., 113, 117–133.597
Snook, N. A. and M. Xue, 2008: Effects of microphysical drop size distribution on tornadoge-598
nesis in supercell thunderstorms. Geophys. Res. Lett., 35, L24803, doi: 10.1029/2008GL035866.599
Spahn, J. E. 1976: The airborne hail disdrometer: An analysis of its 1975 performance. Rep.600
76-13, Institute of Atmospheric Sciences, South Dakota School of Mines and Technology,601
Rapid City, SD, 65 pp.602
Stoelinga, M. T. 2005: Simulated equivalent reflectivity factor as currently formulated in603
RIP: Description and possible improvements, Read/Interpolate/Plot implementation604
document.605
25
Storer, R. L. S. C. van den Heever, and G. L. Stephens, 2010: Modeling aerosol impacts on606
convective storms in different environments. J. Atmos. Sci., 67, 3904–3915.607
Twohy, C. H. P. J. DeMott, K. A. Pratt, R. Subramanian, G. L. Kok, S. M. Murphy, T.608
Lersch, A. J. Heymsfield, Z. Wang, K. A. Prather, and J. H. Seinfeld, 2010: Relation-609
ships of biomass-burning aerosols to ice in orographic wave clouds. J. Atmos. Sci., 67,610
2437–2450.611
van den Heever, S. C. and W. R. Cotton, 2004: The impact of hail size on simulated supercell612
storms. J. Atmos. Sci., 61, 1596–1609.613
—–, G. G. Carrio, W. R. Cotton, P. J. DeMott, and A. J. Prenni, 2006: Impacts of nucleating614
aerosol on Florida storms. Part I: Mesoscale simulations. J. Atmos. Sci., 63, 1752–1775.615
—–, and W. R. Cotton, 2007: Urban aerosol impacts on downwind convective storms. J.616
App. Meteor., 46, 828–850.617
Weisman, M. L, 1993: The genesis of severe long-lived bow echoes. J. Atmos. Sci., 50,618
645–670.619
Weisman, M. L. 1992: The role of convectively generated rear-inflow jets in the evolution of620
long-lived mesoconvective systems. J. Atmos. Sci., 49, 1826–1847.621
Yang, M.J. and R. A. Houze Jr. 1995: Sensitivity of squall-line rear inflow to ice microphysics622
and environmental humidity. Mon. Wea. Rev., 123, 3175–3193.623
Zhang, D. L, and K. Gao, 1989: Numerical simulation of an intense squall line during 10-624
11 June 1985 PRE-STORM. Part II: Rear inflow, surface pressure perturbations, and625
stratiform precipitation. Mon. Wea. Rev., 117, 2067–2094.626
26
FIGURE CAPTIONS
Figure 1. WRF-ARW domain. The entire domain was run in 3-km resolution. Later figures will627
show only a subsection of the domain over eastern Oklahoma for space considerations.628
Figure 2. 0000 UTC 13 March 2003 KOUN sounding, modified for use in idealized simulations.629
Figure 3. Graupel particle size distributions for the model runs given in Table 1.630
Figure 4. 0315 UTC (a), 0600 UTC (b), and 0800 UTC (c) 13 March 2003 WSI NOWrad com-631
posite reflectivity data.632
Figure 5. Simulated composite reflectivity data from WRF hail-like (a, c) and graupel-like (b,633
d) case study simulations. 0600 (a, b) and 0800 (c, d) UTC 13 March 2003. Simulated634
reflectivity calculated as in Stoelinga (2005). Thick black lines delineate location of635
cross-sections in subsequent figures.636
Figure 6. Model output from case study hail-like simulation (a, c) and graupel-like simulation637
(b, d) at 0600 (a, b) and 0800 (c, d) UTC 13 March 2003. The cross-section is taken638
along the black lines in Fig. 5; plotted values are averages of 15 km either side of639
the cross-section. Mixing ratios (g kg−1) are of graupel (image), cloud water and ice640
(dashed red, 0.2 g kg−1), rain water (green, 0.5 g kg−1), and snow (thin black, 0.1 g641
kg−1); the thick solid black lines are cold pool potential temperature contours (2 K)642
at 294 K and colder. Thick dashed black line is the melting level.643
Figure 7. Cooling rates (K (5 min)−1) by evaporation (blue image), melting (red, 0.1K (5 min)−1),644
and sublimation (purple, 0.1K (5 min)−1) for the hail-like (a) and graupel-like (b)645
simulations at 0600 UTC 13 March 2003. Thick solid black lines are the simulated646
reflectivity at 25, 40, and 50 dBZ. Thick dashed black line is the melting level.647
27
Figure 8. As in Fig. 6, but with vertical motion (image, cm s−1) and perturbation pressure (black,648
0.5 hPa). Perturbation pressure is calculated by subtracting the model pressure field649
from the MM5 standard atmosphere, as in Stoelinga (2005).650
Figure 9. Simulated composite reflectivity data from WRF idealized hail-like (a, c) and graupel-651
like (b, d) model simulations. 2:00 (a, b) and 4:30 (c, d) simulation time. Simulated652
reflectivity calculated as in Stoelinga (2005). Horizontal scale is km “east” of left edge653
of domain, and “north” of bottom edge of domain.654
Figure 10. Simulated composite reflectivity data from WRF no-graupel (a,b,c) and graupel (d,e,f)655
case study simulations. Columns, from left to right, correspond to times 0445 (a, d),656
0600 (b, e), and 0730 (c, f) UTC 13 March 2003. Simulated reflectivity calculated as657
in Stoelinga (2005). Thick black lines delineate location of cross-sections in subsequent658
figures.659
Figure 11. Mixing ratio cross-section from the no-graupel (left column) and graupel (right column)660
case study simulation, at 0445 (a, b), 0600 (c, d), and 0730 (e, f) UTC 13 March 2003661
from cross-section lines shown in Fig. 10. Values are averages of 15 km either side662
of cross-section. Mixing ratios (g kg−1) are of cloud water and ice (dashed red, 0.2 g663
kg−1), rain water (green, 0.5 g kg−1), snow (thin black, left column 0.5 g kg−1, right664
column 0.2 g kg−1), and graupel (right column, image). The thick solid black lines are665
cold pool potential temperature contours (2 K) at 294 K and colder. The thick dashed666
black line is the melting level.667
Figure 12. Cross-section of vertical motion (image, cm s−1) and perturbation pressure (black, 0.5668
hPa) for the no-graupel (a) and graupel (b) case study simulations at 0730 UTC 13669
March 2003. Perturbation pressure is calculated by subtracting the model pressure670
field from the MM5 standard atmosphere, as in Stoelinga (2005).671
Figure 13. Simulated composite reflectivity data from idealized WRF model simulations WSM5672
28
(a,b,c) and graupel (d,e,f). Columns, from left to right, correspond to times 2:30 (a, d),673
3:30 (b), 3:00 (e), and 4:00 (c, f) simulation time. Simulated reflectivity calculated as674
in Stoelinga (2005). Horizontal scale is km “east” of left edge of domain, and “north”675
of bottom edge of domain.676
29
FIGURES
Figure 1: WRF-ARW domain. The entire domain was run in 3-km resolution. Later figureswill show only a subsection of the domain over eastern Oklahoma for space considerations.
30
Figure 2: 0000 UTC 13 March 2003 KOUN sounding, modified for use in the idealizedsimulations.
31
Figure 3: Graupel particle size distributions for the model runs given in Table 1.
32
Figure 4: 0315 UTC (a), 0600 UTC (b), and 0800 UTC (c) 13 March 2003 WSI NOWradcomposite reflectivity data.
33
Figure 5: Simulated composite reflectivity data from WRF hail-like (a, c) and graupel-like(b, d) case study simulations. 0600 (a, b) and 0800 (c, d) UTC 13 March 2003. Simulatedreflectivity calculated as in Stoelinga (2005). Thick black lines delineate location of cross-sections in subsequent figures.
34
Figure 6: Model output from case study hail-like simulation (a, c) and graupel-like simulation(b, d) at 0600 (a, b) and 0800 (c, d) UTC 13 March 2003. The cross-section is taken alongthe black lines in Fig. 5; plotted values are averages of 15 km either side of the cross-section.Mixing ratios (g kg−1) are of graupel (image), cloud water and ice (dashed red, 0.2 g kg−1),rain water (green, 0.5 g kg−1), and snow (thin black, 0.1 g kg−1); the thick solid black linesare cold pool potential temperature contours (2 K) at 294 K and colder. Thick dashed blackline is the melting level.
35
Figure 7: Cooling rates (K (5 min)−1) by evaporation (blue image), melting (red, 0.1K (5min)−1), and sublimation (purple, 0.1K (5 min)−1) for the hail-like (a) and graupel-like (b)simulations at 0600 UTC 13 March 2003. Thick solid black lines are the simulated reflectivityat 25, 40, and 50 dBZ. Thick dashed black line is the melting level.
36
Figure 8: As in Fig. 6, but with vertical motion (image, cm s−1) and perturbation pressure(black, 0.5 hPa). Perturbation pressure is calculated by subtracting the model pressure fieldfrom the MM5 standard atmosphere, as in Stoelinga (2005).
37
Figure 9: Simulated composite reflectivity data from WRF idealized hail-like (a, c) andgraupel-like (b, d) model simulations. 2:00 (a, b) and 4:30 (c, d) simulation time. Simulatedreflectivity calculated as in Stoelinga (2005). Horizontal scale is km “east” of left edge ofdomain, and “north” of bottom edge of domain.
38
Figure 10: Simulated composite reflectivity data from WRF no-graupel (a,b,c) and graupel(d,e,f) case study simulations. Columns, from left to right, correspond to times 0445 (a,d), 0600 (b, e), and 0730 (c, f) UTC 13 March 2003. Simulated reflectivity calculated as inStoelinga (2005). Thick black lines delineate location of cross-sections in subsequent figures.
39
Figure 11: Mixing ratio cross-section from the no-graupel (left column) and graupel (rightcolumn) case study simulation, at 0445 (a, b), 0600 (c, d), and 0730 (e, f) UTC 13 March2003 from cross-section lines shown in Fig. 10. Values are averages of 15 km either side ofcross-section. Mixing ratios (g kg−1) are of cloud water and ice (dashed red, 0.2 g kg−1),rain water (green, 0.5 g kg−1), snow (thin black, left column 0.5 g kg−1, right column 0.2 gkg−1), and graupel (right column, image). The thick solid black lines are cold pool potentialtemperature contours (2 K) at 294 K and colder. The thick dashed black line is the meltinglevel.
40
Figure 12: Cross-section of vertical motion (image, cm s−1) and perturbation pressure (black,0.5 hPa) for the no-graupel (a) and graupel (b) case study simulations at 0730 UTC 13 March2003. Perturbation pressure is calculated by subtracting the model pressure field from theMM5 standard atmosphere, as in Stoelinga (2005).
41
Figure 13: Simulated composite reflectivity data from idealized WRF model no-graupel(a,b,c) and graupel (d,e,f) simulations. Columns, from left to right, correspond to times 2:30(a, d), 3:30 (b), 3:00 (e), and 4:00 (c, f) simulation time. Simulated reflectivity calculatedas in Stoelinga (2005). Horizontal scale is km “east” of left edge of domain, and “north” ofbottom edge of domain.
42
TABLES
Table 1: Names of model simulations with the intercept value (n0G) and density (ρG) used.677
These range from the simulation with the most “hail-like” graupel class to the most “graupel-678
like”. The default values in WSM6 and WDM6 are an intercept value of 4x106 m−4, and679
density of 500 kg m−3.680
Simulation Microphysics Intercept (n0G) Density (ρG)name scheme m−4 kg m−3
“hail-like’ WSM6 4x102 900N104D900 WSM6 4x104 900N104D700 WSM6 4x104 700N106D500 WSM6 4x106 500
“gruapel-like” WSM6 4x106 300“no-graupel” WDM5/WSM5 — —
“graupel” WDM6/WSM6 4x106 500
43